Automatic Animal Specie Identification Using Bioacoustics Characterization and Modeling

Animal sounds exhibit differences often clear to the human hearing, allowing humans to recognize species through sound only. It is possible for machines to perform this task in a computationally efficient way with high performance. In this paper, we propose to recognize the species of an animal using sound characteristics. The emitted sound is submitted to Mel-Frequency Cepstral Coding (MFCC) and then to Gaussian Mixture Model (GMM) classification to identify the species. Evaluations conducted on the system with farm animal sounds study the effects of the presence of noises and signal duration. They demonstrate satisfying performance with promising prospects as recognition rates increase with the studied signal durations and with less noise and can reach 95%.